State-Complexity Reduction in MLSD Receivers for Optical Communications With Direct Photodetection

Abstract

Abstract-We investigate the impact of state-complexity reduction on the performance of maximum likelihood sequence detection (MLSD) receivers for direct-photodetection long-haul optical communication systems affected by uncompensated chromatic dispersion (CD). We directly compare two possible approaches: (i) detection through a simple "brute-force" state-complexity reduction strategy and (ii) a more structured reduced-state sequence detection (RSSD) strategy. The performance of both state-complexity reduction techniques is evaluated considering two realistic optical transmission schemes, based on on-off keying (OOK) and optical duobinary modulation (ODBM), respectively. The detection algorithms are characterized considering the impact of the timing offset, the quantization scheme, and the amount of uncompensated CD. As one would expect, for a given number of states in MLSD receivers, the schemes based on RSSD exhibit better performance with respect to those based on simple brute-force state-complexity reduction. However, we show that MLSD schemes based on the use of brute-force state-complexity reduction are characterized by a better complexity/performance trade-off for low/medium CD values, whereas RSSD is the best choice for high CD values.Index Terms-Maximum likelihood sequence detection (MLSD), on-off keying (OOK), optical duobinary modulation (ODBM), optical fiber communications, reduced-state sequence detection (RSSD).

Reduced-State MLSD Based on Volterra Kernels for Square-Law Detected Multipath Channels

“…Consequently, the log likelihood function becomes complicated and closed form expression is not feasible. A straightforward way to implement Viterbi algorithm is to find the (, ) kk r  b functions for all combination of k b , which is 2 M , as proposed in [8].…”

“…Note that the branch metric in Eq. ( 21) is explicit and structured unlike that of reduced-state MLSD equalizers in [8]. Without this Volterra kernel model, L past bits should be separately determined.…”

“…This distortion can be modeld as a sqaure law detection of multipath signals. Much work has been carried out in applying various equalization techniques utilized in wireless communication systems to optical communication systems [1][2][3][4] [5][6] [7] [8]. The fundamental limitaion is that the most studies in wireless communication have focused on linear equalization, while the distortin due to sqaure law detection and multipath is nonlinear.…”

“…In optical communications systems, due to the nonlinearity of the photo detector, a Viterbi algorithm based on a linear channel model is not feasible. Hence, existing reduced-state MLSD techniques for optical systems rely on a brute force truncation [7] or complicated direct computation of branch metrics for all the reduced states reflecting the state reduction [8].…”

Performance of Euclidean-Metric MLSD Receiver in the Presence of Channel Mismatch Caused by Nongaussian Noise

“…At this error rate, computer simulation alone is not a viable tool for performance evaluation; thus, some form of theoretical analysis is required. The error probability of EM-MLSD over dispersive channels has been widely investigated in the literature [2], [5], [6], [7], [8]. For example, [7] investigates the performance of MLSD for generalized nongaussian channels, while a theory of the error rate of MLSD in Gaussian channels is proposed in [5].…”